Authenticating Digital and Mobile Evidence in the Age of AI: Is Rule 901 Ready?

Generative artificial intelligence can now produce images, video, and audio that even trained observers struggle to distinguish from genuine recordings. For litigators, that capability lands on a question as old as photography itself: how does a court decide that a piece of evidence is what its proponent says it is? The answer, for now, still runs through Federal Rule of Evidence 901, and whether the rule should change remains an open question for rule-makers, while the discipline that parties and their experts bring to applying the existing framework matters regardless of how that question is resolved.

The existing authentication framework

Authentication under Fed. R. Evid. 901(a) sets a deliberately modest bar. The proponent need only "produce evidence sufficient to support a finding that the item is what the proponent claims it is." This is a question of conditional relevance: a text message matters only if it is, in fact, the message it purports to be. The judge does not decide authenticity. Under Fed. R. Evid. 104(b), the court asks only whether a reasonable juror could find the item genuine; the trier of fact then weighs it. Rule 901 is meant to screen out items that plainly are not what they claim, not to resolve genuine factual disputes about provenance.

Fed. R. Evid. 901(b) supplies a non-exhaustive menu of methods. A witness with personal knowledge may authenticate under 901(b)(1); an expert or the trier of fact may compare a questioned item against an authenticated specimen under 901(b)(3); distinctive characteristics and surrounding circumstances support 901(b)(4); voice identification falls under 901(b)(5); and evidence describing a process or system that produces an accurate result authenticates under 901(b)(9), the provision courts routinely use for surveillance footage. These illustrations work in combination, and a separate track, Fed. R. Evid. 902(13)-(14), allows certain electronic records and copies authenticated by hash value to be self-authenticating through certification.

How courts authenticate phone, screenshot, and video evidence today

Mobile and digital evidence has strained this framework less than some predicted. When the Advisory Committee on Evidence Rules studied special rules for emails, texts, and social media in 2014, it declined to draft them, finding the existing 901(b) illustrations adaptable, bespoke rules duplicative of distinctive-characteristics analysis, and any narrow rule at risk of obsolescence given a multi-year rulemaking cycle. Courts have largely vindicated that judgment, applying Rule 901 to text messages, Facebook content, and screenshots through authorship cues, account familiarity, reply patterns, and metadata. As one decision authenticating social media and text messages observed, conclusive proof of authorship is not required; corroborating circumstances can carry the showing even where the witness cannot be certain who typed the words.

Critically, courts have refused to let a bare denial defeat authentication. Just as a generic "my account was hacked" assertion does not exclude electronically stored information as a matter of course, a generic objection should not, without more, keep otherwise-foundationed evidence from the jury. That track record is the strongest argument that the framework can absorb the next wave of challenges.

The pressure deepfakes and generative AI add

Generative models complicate this picture, though not catastrophically. Because synthetic media can be highly realistic, fabricated items may clear Rule 901's low threshold more easily than crude forgeries did. Detection is genuinely hard: adversarial training refines models to evade scrutiny, and AI elements can be spliced into otherwise-authentic files so that metadata appears ordinary. A witness who reviews manipulated footage may even internalize false details, eroding the reliability of 901(b)(1) testimony.

There is also the inverse risk, the "liar's dividend": as the public grows aware of deepfakes, a party can falsely brand genuine evidence as synthetic. In Huang v. Tesla, counsel suggested a years-old public recording of Elon Musk might be a deepfake; the court rejected a position that would let public figures disown anything they had actually said. In Mendones v. Cushman & Wakefield, Inc., a California trial court suspected summary-judgment exhibits were AI-generated, ordered production of detailed file metadata, analyzed inconsistencies in lighting and motion, and imposed terminating sanctions. And in Hohsfield v. Staffieri, a court weighed a deepfake allegation against the totality of the circumstances and found it unpersuasive. The early lesson is consistent: courts that ask the right questions are catching both fabrications and frivolous fakery claims.

The case for, and against, amending Rule 901

Two leading proposals would amend Rule 901. One would tighten the process-or-system illustration for AI-generated items and add a new Rule 901(c) shifting the burden once a challenger shows, by a preponderance, that evidence is more likely than not fabricated or altered, after which admissibility turns on a probative-versus-prejudicial balancing. A second proposal would likewise require a threshold showing of manipulation, then demand that the proponent both authenticate under 901(b) and prove reliability, with the judge deciding admissibility under Fed. R. Evid. 104(a) rather than leaving authenticity to the jury. Each is designed to deter frivolous "it's a deepfake" objections while giving courts a concrete standard.

Skeptics counter that courts have absorbed doctored photographs and every prior "truth-subversive" technology without rewriting the rules, that rising authentication costs do not justify amendment, and that a rule drafted today may be outdated before it takes effect. The Advisory Committee has, for now, taken a wait-and-see posture, reasoning that existing tools, watermarks, and hash fingerprints may suffice and that real case data should drive any change. Under that approach, the rule's text stays fixed while judges and counsel adapt their practice; the Advisory Committee has indicated it will await real-world case experience before considering any amendment.

The forensic expert's role in provenance and challenge

Whether or not Rule 901 changes, contested AI-evidence questions will increasingly require qualified forensic analysis, because the realism of modern synthesis often exceeds what a judge or jury can assess unaided. A forensic examiner supports authentication, or a defensible challenge, by working the provenance chain rather than relying on gut impressions. That work typically includes:

• Recovering and preserving original files and full metadata from the source device, not downstream copies that strip provenance.
• Analyzing container structure, codec and encoding artifacts, and timestamps for signs of generation or post-capture editing.
• Comparing questioned media against known-authentic exemplars under 901(b)(3), and documenting hash values to support 902(13)-(14) certification where available.
• Distinguishing benign editing (compression, format conversion) from indicia of fabrication, and stating the limits of what current tools can and cannot establish.

Equally important is candor about uncertainty. A credible expert does not guarantee that media is "real" or "fake"; the expert explains methodology, reproducibility, and the boundaries of detection so the court can perform its gatekeeping function. That discipline matters whether the expert is laying a foundation, conducting an independent review of an opposing forensic report, or preparing reports, declarations, and testimony that will be tested under Fed. R. Evid. 104.

Practical takeaways for counsel

To date, courts have continued to apply Rule 901 to this evidence without adopting a new rule. Rule 901's flexibility, combined with the self-authentication path of 902(13)-(14) and the judge's screening role under 104, gives courts the tools to admit genuine digital and mobile evidence and to scrutinize suspect items, so long as the parties supply real foundation and real challenges. Counsel should treat metadata and chain of custody as first-class issues from preservation forward, resist reflexive deepfake objections that invite Rule 11 exposure, and retain forensic expertise early where provenance is genuinely contested. The right time to address authenticity is at case intake, not on the eve of trial.

Adapted from Law & Forensics continuing-legal-education and seminar materials (2025–2026). This article is general information for attorneys and is not legal advice; it does not create an attorney-client, expert, or consulting relationship.